AAAI 2021 Papers

When exposed to an item in a recommender system, a user may consume it (known as success exposure) or neglect it (known as failure exposure). The recently proposed methods that consider both success and failure exposure merely regard failure exposure as a constant prior, thus being capable of neither modeling various user behavior nor adapting to overdispersed data. In this paper, we propose a novel model, hierarchical negative binomial factorization, which models data dispersion via a hierarchical Bayesian structure, thus alleviating the effect of data overdispersion to help with performance gain for recommendation. Moreover, we factorize the dispersion of zero entries approximately into two low-rank matrices, thus reducing the updating time linear to the number of nonzero entries. The experiment shows that the proposed model outperforms state-of-the-art Poisson-based methods merely with a slight loss of inference speed.

Current multiple kernel clustering algorithms compute a partition with the consensus kernel or graph learned from the pre-specified ones, while the emerging late fusion methods firstly construct multiple partitions from each kernel separately, and then obtain a consensus one with them. However, both of them directly distill the clustering information from kernels or graphs to partition matrices, where the sudden dimension drop would result in loss of advantageous details for clustering. In this paper, we provide a brief insight of the aforementioned issue and propose a hierarchical approach to perform clustering while preserving advantageous details maximumly. Specifically, we gradually group samples into fewer clusters, together with generating a sequence of intermediary matrices of descending sizes. The consensus partition with is simultaneously learned and conversely guides the construction of intermediary matrices. Nevertheless, this cyclic process is modeled into an unified objective and an alternative algorithm is designed to solve it. In addition, the proposed method is validated and compared with other representative multiple kernel clustering algorithms on benchmark datasets, demonstrating state-of-the-art performance by a large margin.

Hierarchically constructing micro (i.e., intra-sentence or inter-sentence) discourse structure trees using explicit boundaries (e.g., sentence and paragraph boundaries) has been proved to be an effective strategy. However, it is difficult to apply this strategy to document-level macro (i.e., inter-paragraph) discourse parsing, the more challenging task, due to the lack of explicit boundaries at the higher level. To alleviate this issue, we introduce a topic segmentation mechanism to detect implicit topic boundaries and then help the document-level macro discourse parser to construct better discourse trees hierarchically. In particular, our parser first splits a document into several sections using the topic boundaries that the topic segmentation detects. Then it builds a smaller and more accurate discourse sub-tree in each section and sequentially forms a whole tree for a document. The experimental results on both Chinese MCDTB and English RST-DT show that our proposed method outperforms the state-of-the-art baselines significantly.

Deep learning based human parsing methods usually require a large amount of training data to reach high performance. However, it is costly and time-consuming to obtain manually annotated high quality labels for a large scale dataset. To alleviate annotation efforts, we propose a new semi-supervised human parsing method for which we only need a small number of labels for training. First, we generate high quality pseudo labels on unlabeled images using a hierarchical information passing network (HIPN), which reasons human part segmentation in a coarse to fine manner. Furthermore, we develop a noise-tolerant hybrid learning method, which takes advantage of positive and negative learning to better handle noisy pseudo labels. When evaluated on standard human parsing benchmarks, our HIPN achieves a new state-of-the-art performance. Moreover, our noise-tolerant hybrid learning method further improves the performance and outperforms the state-of-the-art semi-supervised method (i.e. GRN) by 4.47 points w.r.t mIoU on the LIP dataset.

Traffic forecasting is attracting considerable interest due to its widespread application in intelligent transportation systems. Given the complex and dynamic traffic data, many methods focus on how to establish a spatial-temporal model to express the non-stationary traffic patterns. Recently, the latest Graph Convolution Network (GCN) has been introduced to learn spatial features while the time neural networks are used to learn temporal features. These GCN based methods obtain state-of-the-art performance. However, the current GCN based methods ignore the natural hierarchical structure of traffic systems which is composed of the micro layers of road networks and the macro layers of region networks, in which the nodes are obtained through pooling method and could include some hot traffic regions such as downtown and CBD etc., while the current GCN is only applied on the micro graph of road networks. In this paper, we propose a novel Hierarchical Graph Convolution Networks (HGCN) for traffic forecasting by operating on both the micro and macro traffic graphs. The proposed method is evaluated on two complex city traffic speed datasets. Compared to the latest GCN based methods like Graph WaveNet, the proposed HGCN gets higher traffic forecasting precision with lower computational cost.The website of the code is https://github.com/guokan987/HGCN.git.

Graph Neural Networks (GNNs) draw their strength from explicitly modeling the topological information of structured data. However, existing GNNs suffer from limited capability in capturing the hierarchical graph representation which plays an important role in graph classification. In this paper, we innovatively propose hierarchical graph capsule network (HGCN) that can jointly learn node embeddings and extract graph hierarchies. Specifically, disentangled graph capsules are established by identifying heterogeneous factors underlying each node, such that their instantiation parameters represent different properties of the same entity. To learn the hierarchical representation, HGCN characterizes the part-whole relationship between lower-level capsules (part) and higher-level capsules (whole) by explicitly considering the structure information among the parts. Experimental studies demonstrate the effectiveness of HGCN and the contribution of each component. Code: https://github.com/uta-smile/HGCN

Text coherence plays a key role in document quality assessment. Most existing text coherence methods only focus on similarity of adjacent sentences. However, local coherence exists in sentences with broader contexts and diverse rhetoric relations, rather than just adjacent sentences similarity. Besides, the highlevel text coherence is also an important aspect of document quality. To this end, we propose a hierarchical coherence model for document quality assessment. In our model, we implement a local attention mechanism to capture the location semantics, bilinear tensor layer for measure coherence and max-coherence pooling for acquiring high-level coherence. We evaluate the proposed method on two realistic tasks: news quality judgement and automated essay scoring. Experimental results demonstrate the validity and superiority of our work.

Given limited handwriting scripts, humans can easily visualize (or imagine) what the handwritten words/texts would look like with other arbitrary textual contents. Moreover, a person also is able to imitate the handwriting styles of provided reference samples. Humans can do such hallucinations, perhaps because they can learn to disentangle the calligraphic styles and textual contents from given handwriting scripts. However, computers cannot study to do such flexible handwriting imitation with existing techniques. In this paper, we propose a novel handwriting imitation generative adversarial network (HiGAN) to mimic such hallucinations. Specifically, HiGAN can generate variable-length handwritten words/texts conditioned on arbitrary textual contents, which are unconstrained to any predefined corpus or out-of-vocabulary words. Moreover, HiGAN can flexibly control the handwriting styles of synthetic images by disentangling calligraphic styles from the reference samples. Experiments on handwriting benchmarks validate our superiority in terms of visual quality and scalability when comparing to the state-of-the-art methods for handwritten word/text synthesis. The code and pre-trained models can be found at https://github.com/ganji15/HiGAN.

Although Markov chain Monte Carlo (MCMC) is useful for generating samples from the posterior distribution, it often suffers from intractability when dealing with large-scale datasets. To address this issue, we propose Hierarchical Initialized Alternating Back-propagation (HiABP) for efficient Bayesian inference. Especially, we endow Alternating Backpropagation (ABP) method with a well-designed initializer and hierarchical structure, composing the pipeline of Initializing, Improving, and Learning back-propagation. It saves much time for the generative model to initialize the latent variable by constraining a sampler to be close to the true posterior distribution. The initialized latent variable is then improved significantly by an MCMC sampler. Thus the proposed method has the strengths of both methods, i.e., the effectiveness of MCMC and the efficiency of variational inference. Experimental results validate our framework can outperform other popular deep generative models in modeling natural images and learning from incomplete data. We further demonstrate the unsupervised disentanglement of hierarchical latent representation with controllable image synthesis.

Heterogeneous Graph Neural Networks (HGNNs) have drawn increasing attention in recent years and achieved outstanding performance in many tasks. The success of the existing HGNNs relies on one fundamental assumption, i.e., the original heterogeneous graph structure is reliable. However, this assumption is usually unrealistic, since the heterogeneous graph in reality is inevitably noisy or incomplete. Therefore, it is vital to learn the heterogeneous graph structure for HGNNs rather than rely only on the raw graph structure. In light of this, we make the first attempt towards learning an optimal heterogeneous graph structure for HGNNs and propose a novel framework HGSL, which jointly performs Heterogeneous Graph Structure Learning and GNN parameters learning for classification task. Different from traditional GSL on homogeneous graph, considering the heterogeneity of different relations in heterogeneous graph, HGSL generates each relation subgraph independently. Specifically, in each generated relation subgraph, HGSL not only considers the feature similarity by generating feature similarity graph, but also considers the complex heterogeneous interactions in features and semantics by generating feature propagation graph and semantic graph. Then, these graphs are fused to a learned heterogeneous graph and optimized together with a GNN towards classification objective. Extensive experiments on real-world graphs demonstrate that the proposed framework significantly outperforms the state-of-the-art methods.

Lexical cohesion is a fundamental mechanism for text which requires a pair of words to be interpreted as a certain type of lexical relation (e.g., similarity) to understand a coherent context; we refer to such relations as the contextual lexical relation. However, work on lexical cohesion has not modeled context comprehensively in considering lexical relations due to the lack of linguistic resources. In this paper, we take initial steps to address contextual lexical relations by focusing on the contrast relation, as it is a well-known relation though it is more subtle and relatively less resourced. We present a corpus named Cont 2 Lex to make Contextual Lexical Contrast Recognition a computationally feasible task. We benchmark this task with widely-adopted semantic representations; we discover that contextual embeddings (e.g. BERT) generally outperform static embeddings (e.g. Glove), but barely go beyond 70% in accuracy performance. In addition, we find that all embeddings perform better when CLC occurs within the same sentence, suggesting possible limitations of current computational coherence models. Another intriguing discovery is the improvement of BERT in CLC is largely attributed to its modeling of CLC word pairs co-occurring with other word repetitions. Such observations imply that the progress made in lexical coherence modeling remains relatively primitive even for semantic representations such as BERT that have been empowering numerous standard NLP tasks to approach human benchmarks. Through presenting our corpus and benchmark, we attempt to seed initial discussions and endeavors in advancing semantic representations from modeling syntactic and semantic levels to coherence and discourse levels.

Multi-exit architectures, in which a sequence of intermediate classifiers are introduced at different depths of the feature layers, perform adaptive computation by early exiting ``easy" samples to speed up the inference. In this paper, a novel Harmonized Dense Knowledge Distillation (HDKD) training method for multi-exit architecture is designed to encourage each exit to flexibly learn from all its later exits. In particular, a general dense knowledge distillation training objective is proposed to incorporate all possible beneficial supervision information for multi-exit learning, where a harmonized weighting scheme is designed for the multi-objective optimization problem consisting of multi-exit classification loss and dense distillation loss. A bilevel optimization algorithm is introduced for alternatively updating the weights of multiple objectives and the multi-exit network parameters. Specifically, the loss weighting parameters are optimized with respect to its performance on validation set by gradient descent. Experiments on CIFAR100 and ImageNet show that the HDKD strategy harmoniously improves the performance of the state-of-the-art multi-exit neural networks. Moreover, this method does not require within architecture modifications and can be effectively combined with other previously-proposed training techniques and further boosts the performance.

Hand gestures play a dominant role in the expression of sign language. Current deep-learning based video sign language recognition (SLR) methods usually follow a data-driven paradigm under the supervision of the category label. However, those methods suffer limited interpretability and may encounter the overfitting issue due to limited sign data sources. In this paper, we introduce the hand prior and propose a new hand-model-aware framework for isolated SLR with the modeling hand as the intermediate representation. We first transform the cropped hand sequence into the latent semantic feature. Then the hand model introduces the hand prior and provides a mapping from the semantic feature to the compact hand pose representation. Finally, the inference module enhances the spatio-temporal pose representation and performs the final recognition. Due to the lack of annotation on the hand pose under current sign language datasets, we further guide its learning by utilizing multiple weakly-supervised losses to constrain its spatial and temporal consistency. To validate the effectiveness of our method, we perform extensive experiments on four benchmark datasets, including NMFs-CSL, SLR500, MSASL and WLASL. Experimental results demonstrate that our method achieves state-of-the-art performance on all four popular benchmarks with a notable margin.

Self-supervised learning shows great potential in monocular depth estimation, using image sequences as the only source of supervision. Although people try to use the high-resolution image for depth estimation, the accuracy of prediction has not been significantly improved. In this work, we find the core reason comes from the inaccurate depth estimation in large gradient regions, making the bilinear interpolation error gradually disappear as the resolution increases. To obtain more accurate depth estimation in large gradient regions, it is necessary to obtain high-resolution features with spatial and semantic information. Therefore, we present an improved DepthNet, HR-Depth, with two effective strategies: (1) re-design the skip-connection in DepthNet to get better high-resolution features and (2) propose feature fusion Squeeze-and-Excitation(fSE) module to fuse feature more efficiently. Using Resnet-18 as the encoder, HR-Depth surpasses all previous state-of-the-art(SoTA) methods with the least parameters at both high and low resolution. Moreover, previous SoTA methods are based on fairly complex and deep networks with a mass of parameters which limits their real applications. Thus we also construct a lightweight network which uses MobileNetV3 as encoder. Experiments show that the lightweight network can perform on par with many large models like Monodepth2 at high-resolution with only20%parameters. All codes and models will be available at https://github.com/shawLyu/HR-Depth.

Automatically solving math word problems is a crucial task for exploring the intelligence levels of machines in the general AI domain. It is highly challenging since it requires not only natural language understanding but also mathematical expression inference. Existing solutions usually explore sequence-to-sequence models to generate expressions, where the problems are simply encoded sequentially. However, such models are generally far from enough for understanding problems as similar to humans and lead to incorrect answers. To this end, in this paper, we propose a novel Hierarchical Math Solver (HMS) to make deep understanding and exploitation of problems. In problem understanding, imitating human reading habits, we propose a hierarchical word-clause-problem encoder. Specifically, we first split each problem into several clauses and learn problem semantics from the local clause level to the global problem level. Then, in clause understanding, we propose a dependency-based module to enhance clause semantics with the dependency structure of the problem. Next, in expression inference, we propose a novel tree-based decoder to generate the mathematical expression for the answer. In the decoder, we apply a hierarchical attention mechanism to enhance the problem semantics with context from different levels, and a pointer-generator network to guide the model to copy existing information and infer extra knowledge. Extensive experimental results on two widely used datasets demonstrate that HMS achieves not only better answers but also more reasonable inference.

Many recent invertible neural architectures are based on coupling block designs where variables are divided in two subsets which serve as inputs of an easily invertible (usually affine) triangular transformation. While such a transformation is invertible, its Jacobian is very sparse and thus may lack expressiveness. This work presents a simple remedy by noting that subdivision and (affine) coupling can be repeated recursively within the resulting subsets, leading to an efficiently invertible block with dense, triangular Jacobian. By formulating our recursive coupling scheme via a hierarchical architecture, HINT allows sampling from a joint distribution p(y,x) and the corresponding posterior p(x|y) using a single invertible network. We evaluate our method on some standard data sets and benchmark its full power for density estimation and Bayesian inference on a novel data set of 2D shapes in Fourier parameterization, which enables consistent visualization of samples for different dimensionalities.

Argument structure elaborates the relation among claims and premises. Previous works in persuasiveness prediction do not consider this relation in their architectures. To take argument structure information into account, this paper proposes an approach to persuasiveness prediction with a novel graph-based neural network model, called heterogeneous argument attention network (HARGAN). By jointly training on the persuasiveness and stance of the replies, our model achieves the state-of-the-art performance on the ChangeMyView (CMV) dataset for the persuasiveness prediction task. Experimental results show that the graph setting enables our model to aggregate information across multiple paragraphs effectively. In the meanwhile, our stance prediction auxiliary task enables our model to identify the viewpoint of each party, and helps our model perform better on the persuasiveness prediction.

Non-autoregressive neural machine translation (NAT) generates each target word in parallel and has achieved promising inference acceleration. However, existing NAT models still have a big gap in translation quality compared to autoregressive neural machine translation models due to the multimodality problem: the target words may come from multiple feasible translations. To address this problem, we propose a novel NAT framework ReorderNAT which explicitly models the reordering information to guide the decoding of NAT. Specially, ReorderNAT utilizes deterministic and non-deterministic decoding strategies that leverage reordering information as a proxy for the final translation to encourage the decoder to choose words belonging to the same translation. Experimental results on various widely-used datasets show that our proposed model achieves better performance compared to most existing NAT models, and even achieves comparable translation quality as autoregressive translation models with a significant speedup.

Acquiring sufficient ground-truth supervision to train deep vi- sual models has been a bottleneck over the years due to the data-hungry nature of deep learning. This is exacerbated in some structured prediction tasks, such as semantic segmen- tation, which requires pixel-level annotations. This work ad- dresses weakly supervised semantic segmentation (WSSS), with the goal of bridging the gap between image-level anno- tations and pixel-level segmentation. We formulate WSSS as a novel group-wise learning task that explicitly models se- mantic dependencies in a group of images to estimate more reliable pseudo ground-truths, which can be used for training more accurate segmentation models. In particular, we devise a graph neural network (GNN) for group-wise semantic min- ing, wherein input images are represented as graph nodes, and the underlying relations between a pair of images are char- acterized by an efficient co-attention mechanism. Moreover, in order to prevent the model from paying excessive atten- tion to common semantics only, we further propose a graph dropout layer, encouraging the model to learn more accurate and complete object responses. The whole network is end-to- end trainable by iterative message passing, which propagates interaction cues over the images to progressively improve the performance. We conduct experiments on the popular PAS- CAL VOC 2012 and COCO benchmarks, and our model yields state-of-the-art performance. Our code is available at: https://github.com/Lixy1997/Group-WSSS.